Imputation of Missing Data Using the Combination of Singular Spectrum Analysis Method and Kalman Filter Algorithm and Comparison with Univariate Imputation Methods

Missing values in time series data are one of the problems that sometimes arise in time series analysis. The more accurate imputation of these values, the better understanding of the structure of the time series will be obtained, and as a result, the recognition of its pattern and the prediction of future values will be more accurate. Therefore, choosing an appropriate method of imputation is an important part of a time series analysis. In this paper we introduce the new missing data imputation method from the singular spectrum analysis procedure, using the Kalman filter algorithm. Then other methods of imputation of missing values in univariate time series are introduced, and will be compared the mentioned methods using simulated data in structural models and real data. The results of the comparison based on the criteria of root mean square error and mean absolute deviations show that the method of imputation of missing values based on singular spectrum analysis approach using the Kalman filter algorithm has a better performance than the other imputation methods and the mode method is the worst.‎‎